要实现所需的输出，最好的办法是使用UDF，它将以行方式工作。

数据准备

s = StringIO("""
category|value
A|4, 5, 6
A|1, 2, 3
B|7, 8, 9
"""
)

df = pd.read_csv(s,delimiter='|')

sparkDF = sql.createDataFrame(df)\
            .withColumn('value',F.split(F.col('value'),',').cast(ArrayType(IntegerType())))

sparkDF.show()

+--------+---------+
|category|    value|
+--------+---------+
|       A|[4, 5, 6]|
|       A|[1, 2, 3]|
|       B|[7, 8, 9]|
+--------+---------+

root
 |-- category: string (nullable = true)
 |-- value: array (nullable = true)
 |    |-- element: integer (containsNull = true)

全要素聚合-自定义项

@F.udf(returnType=ArrayType(IntegerType()))
def custom_arr(inp):
    
    res = np.array(inp)

    return res.sum(axis=0).tolist()

sparkDF = sparkDF.groupBy('category')\
                 .agg(F.collect_list('value').alias('value'))\
                 .withColumn('final_arr',custom_arr(F.col('value')))

sparkDF.show(truncate=False)

+--------+----------------------+---------+
|category|value                 |final_arr|
+--------+----------------------+---------+
|A       |[[4, 5, 6], [1, 2, 3]]|[5, 7, 9]|
|B       |[[7, 8, 9]]           |[7, 8, 9]|
+--------+----------------------+---------+